KR20200114173A - Method, server and computer program for managing natural language processing engines - Google Patents

Abstract

According to one embodiment of the present invention, provided is a method for managing a natural language processing engine, which comprises the steps of: receiving upload data consisting of question data and response data corresponding to the question data; performing an indexing process on the received upload data; training a natural language processing engine based on similar sentences based on the upload data on which the indexing process has been performed; creating a chatbot using the generated natural language processing engine based on the similar sentences; and deriving a response to the query message from the uploaded data on which the indexing process has been performed by using the similar sentence-based natural language processing engine through the generated chatbot.

Description

How to manage natural language processing engines, server and computer programs {METHOD, SERVER AND COMPUTER PROGRAM FOR MANAGING NATURAL LANGUAGE PROCESSING ENGINES}

The present invention relates to a method for managing a natural language processing engine, a server, and a computer program, and more specifically, a chatbot using a similar sentence-based natural language processing engine or a chatbot using a natural language processing model to derive a response to a query message. It relates to a method, a server, and a computer program for managing a natural language processing engine.

IMS (Instant Message Service) refers to a real-time online communication service that allows messages sent by the sender to appear on the screen of the recipient's device immediately without using an e-mail program or web browser. In particular, a service that provides chat, voice, and video calls using a mobile network is called a MIM (Mobile Instant Messenger) service.

Recently, Chatbot or Messenger Bot, which is a robot that chats like a human, is being applied to IMS, and through this, companies and customers automatically provide 1:1 customized information desired by customers through one-on-one conversations within the messenger. It became possible to provide.

In general, a chatbot or a messenger bot can automatically generate a response message by interpreting a message input from a user and provide it to the user. However, chatbots or messenger bots have the disadvantage that they often provide inappropriate answers to questions when a question that cannot be responded mechanically is received.

Korean Registered Patent Publication No. 10-1605430 (published on Aug. 24, 2015)

The present invention is to solve the above-described problems of the prior art, by registering upload data consisting of a set of a query message and a response message corresponding to the query message to a bot builder capable of creating a chatbot, thereby providing upload data and a chatbot platform. We would like to provide a server and computer program, a method for managing a natural language processing engine that allows you to connect and create a chatbot.

A method of managing a natural language processing engine, a server and a computer program that creates a search-type chatbot that uses a similar sentence-based natural language processing engine to provide an answer corresponding to a similar query sentence by referring to a similar query sentence similar to the query message. Want to provide.

A natural language processing engine that does not process a fallback when a user's intention cannot be extracted from a query message in a conventional machine learning-based chatbot, but instead derives a response using a similar sentence-based natural language processing engine. I would like to provide a way to manage the system, a server and a computer program.

When using a natural language processing model composed of a similar sentence-based natural language processing engine and an intention-based natural language processing engine, a response is derived for each of the similar sentence-based natural language processing engine and the intention-based natural language processing engine, and based on the reliability of each derived response. It is intended to provide a method, a server, and a computer program for managing a natural language processing engine capable of providing a response with higher accuracy by providing a response with higher reliability to a user terminal.

As a technical means for achieving the above-described technical problem, registering upload data consisting of question data and response data corresponding to the question data according to an embodiment of the present invention, and indexing the registered upload data. Performing, learning a similar sentence-based natural language processing engine based on uploaded data on which the indexing process has been performed, creating a chatbot using the generated similar sentence-based natural language processing engine, and the generated chatbot. A natural language processing engine management method including the step of deriving a response to a query message from upload data on which the indexing process has been performed using a natural language processing engine based on a similar sentence may be provided.

An upload data registration unit receiving upload data consisting of question data and response data corresponding to the question data according to another embodiment of the present invention, an index unit performing an indexing process on the registered upload data, and the indexing process is performed A similar sentence-based natural language processing engine learning unit for learning a similar sentence-based natural language processing engine based on the uploaded data, a chatbot generating unit generating a chatbot using the generated similar sentence-based natural language processing engine, and the generated chatbot. A management server including a response derivation unit that derives a response to a query message from upload data on which the indexing process has been performed using a similar sentence-based natural language processing engine may be provided.

When executed by a computing device, a computer program according to another embodiment of the present invention registers upload data composed of question data and response data corresponding to the question data, and performs an indexing process on the registered upload data. , Learning a similar sentence-based natural language processing engine based on the uploaded data on which the indexing process is performed, generating a chatbot using the generated similar sentence-based natural language processing engine, and processing the similar sentence-based natural language through the generated chatbot It is possible to provide a computer program stored in a medium including a sequence of instructions for deriving a response to a query message from upload data on which the indexing process has been performed using an engine.

The present invention is to solve the problems of the prior art, and the similar sentence natural language processing engine learned by receiving upload data indexes the uploaded data by referring to the unit word defined and registered in the chatbot platform when deriving a response. By performing a process, searching for a plurality of similar query sentences through an indexing process, calculating the similarity accuracy in a score form from the searched plurality of similar query sentences, and providing a query sentence and response based on the calculated similarity accuracy to the user, A method, a server, and a computer program for managing a natural language processing engine that searches and provides a response to a query message based on a knowledge base may be provided.

In addition, in the case where at least one natural language processing engine (intention-based natural language processing engine and similar sentence-based natural language processing engine) is combined with the natural language processing model of the present invention, the intention-based natural language processing engine cannot extract the user's intention from the query message. , A natural language processing engine that allows responses to be derived from a similar sentence-based natural language processing engine to work complementarily with both the advantages of machine learning or pattern-based chatbots and of knowledge search-based chatbots using uploaded data. How to manage, server and computer programs can be provided.

Further, in the present invention, the threshold level is set for the activation rate between the first processing engine and the second processing engine, determining the processing order of the first processing engine and the second processing engine, and determining the processing order of the first processing engine and the second processing engine. It is possible to provide a method, a server, and a computer program for managing a natural language processing engine that enables the degree of processing by the user to be customized.

In addition, the present invention sets a threshold for an activation rate between the first processing engine and the second processing engine for a sequentially linked processing model composed of different natural language processing engines, so that the accuracy of the response derived through the first processing engine is a threshold value. If the level is exceeded, the response derived through the first processing engine is provided as the final response, and when the accuracy of the response derived through the first processing engine is less than the threshold level, the response derived through the second processing engine is finalized. It is possible to provide a method, a server, and a computer program for managing a natural language processing engine that provides a response as a response and provides a response to a user by complementarily using the first processing engine and the second processing engine.

In addition, the present invention sets a weight level between the first processing engine and the second processing engine for a simultaneous linked processing model composed of different natural language processing engines, thereby providing a higher level of reliability based on the respective reliability derived through each processing engine. It is possible to provide a method, a server, and a computer program for managing a natural language processing engine that enables a response indicating reliability to be provided to a user as a final response.

1 is a block diagram of a natural language processing engine management system according to an embodiment of the present invention.
2 is a configuration diagram of a management server according to an embodiment of the present invention.
3A to 3E are exemplary diagrams for explaining a process of registering upload data according to an embodiment of the present invention.
4A and 4B are exemplary views for explaining a process of learning a similar sentence-based natural language processing engine by selecting a plurality of previously registered updates according to an embodiment of the present invention.
5A and 5B are exemplary diagrams for explaining a process of setting a threshold level for an activation rate according to an embodiment of the present invention.
6A to 6E are exemplary diagrams for explaining a process of setting a derived response to be output in an immediate response type according to an embodiment of the present invention.
7A to 7C are exemplary diagrams for explaining a process of setting a derived response to be output in an answer recommendation type according to an embodiment of the present invention.
8A and 8B are flowcharts of a method of managing a natural language processing engine in a management server according to an embodiment of the present invention.
9A and 9B are exemplary diagrams for explaining a process of deriving a response to a query message by using a natural language processing model in a management server according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and similar reference numerals are assigned to similar parts throughout the specification.

Throughout the specification, when a part is said to be "connected" to another part, this includes not only "directly connected" but also "electrically connected" with another element interposed therebetween. . In addition, when a part "includes" a certain component, it means that other components may be further included rather than excluding other components unless specifically stated to the contrary.

In the present specification, the term "unit" includes a unit realized by hardware, a unit realized by software, and a unit realized using both. Further, one unit may be realized using two or more hardware, or two or more units may be realized using one hardware. Meanwhile,'~ unit' is not meant to be limited to software or hardware, and'~ unit' may be configured to be in an addressable storage medium or configured to reproduce one or more processors. Thus, as an example,'~ unit' refers to components such as software components, object-oriented software components, class components and task components, processes, functions, properties, and procedures. , Subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays and variables. The components and functions provided in the'~ units' may be combined into a smaller number of elements and'~ units', or may be further divided into additional elements and'~ units'. In addition, components and'~ units' may be implemented to play one or more CPUs in a device or a security multimedia card.

The "user terminal" mentioned below may be implemented as a computer or portable terminal that can access a server or other terminal through a network. Here, the computer includes, for example, a notebook equipped with a web browser, a desktop, a laptop, and the like, and the portable terminal is, for example, a wireless communication device that ensures portability and mobility. , Smart phones, tablet PCs, wearable devices, as well as communication modules such as Bluetooth (BLE, Bluetooth Low Energy), NFC, RFID, Ultrasonic, infrared, Wi-Fi, and LiFi. It can include various mounted devices. In addition, "network" refers to a connection structure in which information exchange is possible between each node such as terminals and servers, and includes a local area network (LAN), a wide area network (WAN), and the Internet. (WWW: World Wide Web), wired and wireless data networks, telephone networks, wired and wireless television networks, etc. Examples of wireless data networks include 3G, 4G, 5G, 3rd Generation Partnership Project (3GPP), Long Term Evolution (LTE), World Interoperability for Microwave Access (WIMAX), Wi-Fi, Bluetooth communication, infrared communication, and ultrasound. Communication, Visible Light Communication (VLC), LiFi, etc. are included, but are not limited thereto.

1 is a block diagram of a natural language processing engine management system according to an embodiment of the present invention.

Referring to FIG. 1, the natural language processing engine management system 1 may include a management server 110, a messenger server 120, and a user terminal 130. Here, the management server 110 may perform the same role as a chatbot server that provides an existing chatbot platform, and the natural language processing engine management system 1 of FIG. 1 is only an embodiment of the present invention. The present invention is not limited to the interpretation. That is, according to various embodiments of the present invention, the natural language processing engine management system may be configured differently from that of FIG. 1.

According to an embodiment, the management server 110 may build and manage a natural language processing engine. You can create and manage chatbots based on the built natural language processing engine. The natural language processing engine managed by the management server 110 may include, but is not limited to, a similar sentence-based natural language processing engine and an intention-based natural language processing engine.

The management server 110 may register upload data consisting of question data and response data corresponding to the question data through a user interface for a bot builder capable of generating a chatbot from an administrator. For example, the management server 110 may register upload data from an administrator through a file upload method or an API connection method with the management server 110.

The management server 110 may learn a similar sentence-based natural language processing engine based on the registered upload data. For example, the management server 110 may select at least one or more of a plurality of previously registered upload data, and learn a similar sentence-based natural language processing engine based on the selected upload data.

The management server 110 may generate a chatbot using the generated similar sentence-based natural language processing engine, and derive a response to the query message using the similar sentence-based natural language processing engine through the generated chatbot. For example, the management server 110 may perform an indexing process on the uploaded data, learn a natural language processing engine based on similar sentences based on the uploaded data on which the indexing process has been performed, and process similar sentences based natural language through a chatbot. A response can be derived from the uploaded data on which the indexing process has been performed using the engine.

Here, the chatbot using the similar sentence-based natural language processing engine may operate as a search-type bot that refers to a predetermined similar query sentence for a query message and derives a response corresponding thereto.

In addition, the management server 110 may receive information on a unit word and learn an intention-based natural language processing engine based on this. The management server 110 may generate a chatbot using an intention-based natural language processing engine, and derive a response to a query message using the intention-based natural language processing engine through the generated chatbot.

The unit word may be composed of'@entity' and a parameter to extract and understand the user's intent and the user's intent. The parameter may refer to a parameter for transmitting data for transmission to an external computer/server for an entity extracted from a user's speech. Accordingly, according to the present invention, the chatbot can transmit a required value of a specific action through a parameter. An entity is a kind of data dictionary, and may mean a collection of important representative words for defining user intention in user speech. Entities are, for example,'@singer name','@mood','@sys.address' (address),'@sys.broadcst.name' (broadcaster name),'@sys.city' (overseas city), and Include'@sys.currency.code' (currency code),'@sys.currencyname' (currency name),'@sys.date' (date),'@sys.date.lunar' (lunar date), etc. I can. For example, when receiving a query message "IU Wishing song" from the user terminal 130, "singer - IU 'based on the' @ singer 'it is able to recognize.

Here, the chatbot using the intention-based natural language processing engine may operate as an intention-based derivation type chatbot that determines an intention for a query message and derives a response corresponding thereto.

According to another embodiment, the management server 110 may build a plurality of natural language processing models by combining a similar sentence-based natural language processing engine and an intention-based natural language processing engine, and manage the constructed natural language processing models. In this case, the management server 110 may receive at least one of a plurality of natural language processing models built by an administrator to generate a chatbot. The multiple natural language processing models are'natural language processing model using only natural language processing engine based on similar sentences','natural language processing model using only natural language processing engine based on intention', and'natural language processing engine based on similar sentences and natural language processing engine based on intention'. And a'simultaneous linkage processing model in which a similar sentence-based natural language processing engine and an intention-based natural language processing engine are used simultaneously' may be included. Here, in the case of a natural language processing model based on a complex natural language processing engine using at least two natural language processing engines, a threshold level or a weighting level capable of customizing the processing degree by each engine in the relationship between at least two natural language processing engines Can be set.

The management server 110 may generate a chatbot using a selected natural language processing model from among a plurality of built natural language processing models. A chatbot using a natural language processing model configured to be processed as a sequential linkage processing model can sequentially use a similar sentence-based natural language processing engine and an intention-based natural language processing engine in a preset order to derive a response.

For example, when the sequential linkage processing model is selected, the management server 110 derives a response through the natural language processing engine set as the first processing engine, and when the response is not derived through the first processing engine, the second processing Responses can be derived through the natural language processing engine set as the engine. The management server 110 calculates the accuracy of the response derived through the natural language processing engine set as the first processing engine, and the accuracy of the calculated response is a threshold level for a preset activation rate between the first processing engine and the second processing engine. If exceeds, the derived response can be confirmed as the final response. Otherwise, when the accuracy of the calculated response is less than or equal to the threshold level for the activation rate, the response may be derived through the natural language processing engine set as the second processing engine. Here, the order of operations between the first processing engine and the second processing engine may be set by an administrator.

For another example, a chatbot using a natural language processing model configured to be processed as a simultaneous linkage processing model may use a similar sentence-based natural language processing engine and an intention-based natural language processing engine to derive a response at the same time.

When the simultaneous linkage processing model is selected, the management server 110 derives a response through each of the natural language processing engine set as the first processing engine and the natural language processing engine set as the second processing engine, and calculates the reliability of each derived response. , It is possible to determine the final response among the derived responses based on the calculated reliability of each response. Here, the management server 110 may assign a predetermined weight to the reliability of each derived response, and determine a final response among the derived responses based on the reliability of each response to which the weight is assigned. Through this process, the management server 110 may provide a highly accurate response to the user by providing a highly reliable response among the responses derived to the query message to the user terminal 130.

The management server 110 used in this and other embodiments may receive an inquiry message input from the user terminal 130 from the messenger server 120. At this time, when the management server 110 receives the official account ID and the inquiry message from the messenger server 120, the management server 110 may derive a response to the inquiry message using a chatbot corresponding to the official account ID.

The management server 110 may transmit the derived response to the messenger server 120. At this time, the response transmitted to the messenger server 120 may be transmitted to the user terminal 130.

The messenger server 120 may provide an instant messaging service that enables the user terminal 130 and other users registered as real friends to exchange conversations in a chat room. In addition, the messenger server 120 may provide an instant messaging service that enables the user terminal 130 to exchange conversations with a chatbot or an official account registered as a virtual friend in the messenger app.

For example, when the messenger server 120 receives a chatbot ID and an inquiry message regarding a selected service from the user terminal 130, the messenger server 120 may transmit the received chatbot ID and an inquiry message to the management server 110. For another example, when the messenger server 120 receives an official account ID and an inquiry message for a selected service from the user terminal 130, the received official account ID (or a chatbot ID mapped with the received official account ID) And the received query message may be delivered to the management server 110.

The messenger server 120 may receive a response to an inquiry message derived by the chatbot from the management server 110. The messenger server 120 may transmit the received response to a chat room opened between the user terminal 130 and a chatbot or a chat room opened between the user terminal 130 and an official account.

Friends that can be registered through the messenger app installed on the user terminal 130 may include real friends and virtual friends.

The user of the user terminal 130 may register other users (eg, family members, acquaintances, co-workers, etc.) without a commercial purpose as real friends through the messenger app. The user terminal 130 may provide an instant messaging service to a user by exchanging a conversation with another user registered as a real friend through a chat room through a messenger app installed on the user terminal 130.

The user of the user terminal 130 may register an official account corresponding to a specific service as a virtual friend through the messenger app. The official account may mean an account operated by a specific enterprise or organization to provide a specific service. For example, the official account may provide information on a specific service to the user terminal 130 through a chat room opened between the official account and the user terminal 130. The user terminal 130 may register the corresponding official account as a virtual friend and receive information on a specific service from the official account. On the other hand, even if the official account is not registered as a virtual friend, information on a specific service may be provided.

The official account may use a chatbot to provide information on a specific service provided by a specific company or organization to the user terminal 130. The administrator of the official account creates a chatbot through the user interface for the bot builder provided from the management server 110, and connects the created chatbot with the official account to link information on a specific service (for example, a specific service Can provide a response to a query message related to. In addition, the administrator of the official account can manage the official account and chatbot by matching one-to-one through the user interface for the bot builder, and can be extended to link multiple chatbots with the official account.

In addition, a user of the user terminal 130 may register a chatbot corresponding to a specific service as a virtual friend through a messenger app. For example, a chatbot registered as a virtual friend may provide information on weather services, stock services, and traffic services to the user terminal 130.

Therefore, when registering an official account as a virtual friend, the user terminal 130 may receive a response to a query message through a chatbot linked to the official account. When registering the chatbot itself as a virtual friend, the corresponding chatbot Through this, a response to the inquiry message can be provided.

The user terminal 130 may provide an instant messaging service to a user by exchanging conversations with, for example, a chatbot or an official account registered as a virtual friend in the messenger app through a chat room. In addition, the user terminal 130 may invite a chatbot registered as a virtual friend to a chat room opened between the user terminal 130 and a real friend.

The user terminal 130 may receive one service selected from a user among a plurality of services provided through a messenger app installed on the user terminal 130. For example, the plurality of services may include a service provided by an official account and a service provided by a chatbot. By selecting an official account or chatbot that is a virtual friend, the user can receive the services provided by the official account or the services provided by the chatbot through the chat room.

The user terminal 130 may transmit service account information corresponding to the selected service to the messenger server 120 when one of a plurality of services provided through a messenger app installed on the user terminal 130 is selected by the user. . For example, the service account information may include a chatbot ID (virtual chatbot ID or real chatbot ID) corresponding to the selected service or an official account ID of an official account corresponding to the selected service.

In addition, when the user terminal 130 receives an inquiry message related to a service selected by the user from the user through a chat room, the user terminal 130 may transmit the corresponding inquiry message to the messenger server 120. The service account information and query message may be delivered to the management server 110 through the messenger server 120.

The user terminal 130 may receive a response corresponding to the query message from the messenger server 120 and display the received response through a display. For example, the user terminal 130 may display the received response through the display. For another example, the user terminal 130 may display a predetermined similar query sentence similar to the query message through the display. In this case, the user terminal 130 may receive a selection of one of the predetermined similar query sentences from the user and display a response to the selected similar query sentence.

In the present invention, a chatbot can be easily created by connecting upload data consisting of a set of question data and response data corresponding to the question data held by an administrator or developer with a machine learning or pattern-based chatbot platform. The chatbot generated through this process may operate as a search-type bot that refers to a query similar to a user query message and derives a corresponding response through a similar sentence-based natural language processing engine generated based on upload data. Alternatively, the chatbot does not fallback the user's query message that has not been processed based on machine learning or pattern, and can complement this by searching for a query similar to the user's query and derives and provides a response linked thereto. .

The operation of each component included in the natural language processing engine management system of FIG. 1 will be described in more detail with reference to the following drawings.

2 is a configuration diagram of a management server according to an embodiment of the present invention. 2, the management server 110 includes an upload data registration unit 210, a similar sentence-based natural language processing engine learning unit 220, a natural language processing model management unit 230, a setting unit 240, and a chatbot generation unit 250. ), a response derivation unit 260 and a communication unit 270 may be included. However, the configuration of the management server 110 according to an embodiment of the present invention may be different from that of FIG. 3.

Here, the administrator who manages the management server 110 may access the bot builder unit of the management server 110 through a web or an app, and may create a chatbot through a user interface provided by the bot builder unit.

According to an embodiment, the management server 110 may generate a chatbot using a similar sentence-based processing engine. Hereinafter, a process of generating a chatbot using a similar sentence-based processing engine will be described.

The upload data register 210 may register upload data consisting of question data and response data corresponding to the question data through a user interface for a bot builder capable of generating a chatbot. When the upload data is composed of question data and response data corresponding to the question data, the upload data register 210 identifies the question and answer created by the administrator through data analysis on the question data and response data, and registers the upload data. can do.

For example, the upload data register 210 may register upload data for connection with the chatbot platform through a file upload method or an API connection method with the management server 110. Here, the upload data register 210 may apply a spam filter function to the uploaded data registered from the administrator. For example, the upload data register 210 receives a specific set of words (eg, profanity, profanity, spam, adult, etc.) as dictionary information from the administrator, so that it interacts with the user terminal 130 through a spam filter function. You can increase the response effect.

A process of registering upload data will be described in detail through FIGS. 3A to 3D.

3A to 3E are exemplary diagrams for explaining a process of registering upload data according to an embodiment of the present invention.

3A is an exemplary diagram showing a user interface related to a bot builder according to an embodiment of the present invention. Referring to FIG. 3A, the management server 110 may display a knowledge extension page 302 through a'knowledge +'menu 301 of the main bot builder screen 300. In this case, in order to combine the similar sentence-based natural language processing engine and the chatbot platform, the administrator may set whether to use the similar sentence-based natural language processing engine through on/off of the use button 303.

The administrator may select the file upload button 305 or the API connection button 306 of the knowledge learning setup 304 to connect the uploaded data stored in the database to the chatbot, and input the save button 307. Through this process, an initial setting for learning a natural language processing engine based on similar sentences can be made.

3B is an exemplary diagram illustrating a process of registering upload data through a file upload method according to an embodiment of the present invention. Referring to FIG. 3B, the upload data registration unit 210 receives a spreadsheet-type (CSV, excel, etc.) file upload button 305 among knowledge upload methods from an administrator, and then uploads data through the knowledge file upload button 310. Can be registered. Through this process, the chatbot can learn knowledge based on the registered upload data.

3C and 3D are exemplary diagrams for explaining a process of registering upload data through an API connection method with the management server 110 according to an embodiment of the present invention. Referring to FIG. 3C, the upload data register 210 receives an API (JSON/GET/POST) connection button 306 among knowledge upload methods from an administrator, and then receives upload data through the connection creation button 320. I can. Referring to FIG. 3D, when the upload data register 210 receives the API connection button 306 from the administrator, the target knowledge name 322 of the basic setting 321, the API URL ( 323), header value setting 324, etc. may be input and upload data may be registered.

The upload data registration unit 210 can easily connect the upload data and the bot builder when upload data is stored in a database connected to an existing chatbot platform through this API connection method.

3E is an exemplary diagram showing history information on upload data according to an embodiment of the present invention. Referring to FIG. 3E, the upload data register 210 may integrate and manage the history information 340 of upload data registered by the file upload method 331 or the API connection method 332. At this time, the upload data registration unit 210 displays only the history information of the upload data of the file upload method 331 when the'file upload method 331' is selected from the knowledge upload method 330 from the administrator, and When'API connection method 332' is selected among the knowledge upload methods 330, only the history information of upload data of the API connection method 332 may be displayed. Here, the history information 340 may include version information of uploaded upload data, upload date and time information, and the like.

Returning to FIG. 2 again, the similar sentence-based natural language processing engine learning unit 220 may learn a similar sentence-based natural language processing engine based on registered upload data or upload data on which an indexing process has been performed. For example, the similar sentence-based natural language processing engine learning unit 220 may select at least one or more of a plurality of previously registered upload data, and learn a similar sentence-based natural language processing engine based on the selected upload data. Here, the similar sentence-based natural language processing engine may be added-on in an expanded form within the chatbot platform or may be operated in an embedded form. A process of learning a similar sentence-based natural language processing engine will be described in detail with reference to FIGS. 4A and 4B.

4A and 4B are exemplary views for explaining a process of learning a similar sentence-based natural language processing engine by selecting a plurality of previously registered updates according to an embodiment of the present invention.

4A is an exemplary diagram illustrating a user interface for a bot builder according to an embodiment of the present invention. Referring to FIG. 4A, when a plurality of upload data 400 including a past domain 1, a past domain 2, a past domain 3, and a new domain are previously registered, the administrator may at least one of the plurality of previously registered upload data 400. You can choose more than one.

The similar sentence-based natural language processing engine learning unit 220 selects a past domain 1, a past domain 2, and a new domain among a plurality of previously registered upload data 400, and based on the selected upload data 401, the similar sentence-based natural language You can train the processing engine. Here, the selected upload data 401 may be used as a knowledge parameter to be reflected as a search source when a response is derived.

4B is an exemplary diagram illustrating a modification screen of a natural language processing engine based on a similar sentence according to an embodiment of the present invention. 4B, the similar sentence-based natural language processing engine learning unit 220 responds to question data and question data of upload data included in the similar sentence-based natural language processing engine generated through the'correct knowledge' button 410 Response data 411 is displayed, and correction or deletion may be input from an administrator.

The administrator may input question data and response data corresponding to the question data according to the spreadsheet-type front UI, and merge and modify them with previously registered upload data to be used as a search answer source.

Returning to FIG. 2 again, the response derivation unit 260 may derive a response to the query message using a similar sentence-based natural language processing engine through the generated chatbot. Here, the number of derived responses may be single or plural.

The index unit (not shown) may perform an indexing process on the uploaded data, and refer to a unit word defined and registered in the chatbot platform to perform the indexing process. In this case, the response derivation unit 260 may derive a response from the uploaded data on which the indexing process has been performed using a similar sentence-based natural language processing engine used through a chatbot, and the similar sentence-based natural language processing engine learning unit 220 The indexing process may train a similar sentence-based natural language processing engine based on the uploaded data.

Here, the index unit may perform indexing using information tagged with a unit word defined in the chatbot platform. The indexing unit can reduce the index loading time, which is indispensable for large-capacity uploads, and improves the inconvenience of the pre-indexing structure by performing a quick indexing process through a connected search engine. -time) can provide the advantage of being able to quickly index large amounts of data.

The response derivation unit 260 searches for a similar query sentence similar to the query message using sentence-vector conversion technology (Sent2Vec*, etc.), and extracts the similar query sentence based on similarity in a separate score form. Can be extracted. Here, the sentence-vector conversion technique refers to a technique for measuring similarity based on the distance between vectors by displaying sentences as vectors.

According to another embodiment, the management server 110 may generate a chatbot using a natural language processing model. Hereinafter, a process of generating a chatbot using a natural language processing model will be described.

The natural language processing model management unit 230 may build a plurality of natural language processing models based on at least one of a similar sentence-based natural language processing engine and an intention-based natural language processing engine. In this case, the natural language processing model management unit 230 may receive a selection of at least one of a plurality of natural language processing models constructed to generate a chatbot. The complex natural language processing model consists of'natural language processing model using only natural language processing engine based on similar sentences','natural language processing model using only natural language processing engine based on intention', and'natural language processing engine based on similar sentences and natural language processing engine based on intention'. And a'simultaneous linked processing model in which a similar sentence-based natural language processing engine and an intention-based natural language processing engine are used at the same time'. Since the natural language processing model has a different quality of processing natural language according to the sequential linkage processing model or the simultaneous linkage processing model, the administrator can select the optimal natural language processing model through Iteration Test in the bot builder domain. The natural language processing model can improve the accuracy of a response by using both an intention-based natural language processing engine and a similar sentence-based natural language processing engine.

When constructing a natural language processing model by selecting at least two natural language processing engines, the setting unit 240 may receive a threshold level for a processing degree between the selected two natural language processing engines. For example, the administrator can set the threshold level by adjusting the bar displayed on the activation gauge through the chatpot platform.

The setting unit 240 may receive an output type of a response to be provided to the user terminal 130. Here, N responses may be output based on the degree of similarity, and the setting unit 240 may set the number of outputs from the manager. For example, the setting unit 240 is an immediate-response type for outputting a speech bubble including one response to one query message, or a predetermined similar query sentence extracted from upload data previously registered for one question message. (For example, 3) and a speech balloon including a response to each similar query sentence may be output to be set to one of the answer recommendation types in which a response is finally selected from the user terminal 130.

The chatbot generation unit 250 may generate a chatbot using a selected natural language processing model from among a plurality of built natural language processing models. For example, when the'natural language processing model using only a natural language processing engine based on a similar sentence' is selected, the chatbot generating unit 250 may generate a chatbot that processes a query message using a natural language processing engine based on a similar sentence. If'natural language processing model using only intention-based natural language processing engine' is selected, a chatbot that processes query messages using intent-based natural language processing engine can be created, and'similar sentence-based natural language processing engine and intention-based natural language processing engine' When this sequentially used sequential linkage processing model or'simultaneous linkage processing model in which a similar sentence-based natural language processing engine and an intention-based natural language processing engine are used at the same time' is selected, a similar sentence-based natural language processing engine and an intention-based natural language processing engine You can create a chatbot that processes query messages by using all of them.

The response derivation unit 260 may derive a response to the query message using the natural language processing model selected through the generated chatbot.

For example, in the case of the natural language processing model selected as the sequential linkage processing model, the chatbot derives responses by sequentially using the similar sentence-based natural language processing engine and the intention-based natural language processing engine in a preset order using the sequential linkage processing model. can do. When the sequential linkage processing model is selected, a response is derived through the natural language processing engine set as the first processing engine, and if the response is not derived through the first processing engine, the response is sent through the natural language processing engine set as the second processing engine. Can be derived. Here, the first processing engine may be an intention-based natural language processing engine, and the second processing engine may be set as a similar sentence-based natural language processing engine, but the order is not limited thereto. The response derivation unit 260 calculates the accuracy of the response derived through the natural language processing engine set as the first processing engine, and the accuracy of the calculated response is a threshold value for the activation rate preset between the first processing engine and the second processing engine. If the level is exceeded, the derived response can be determined as the final response. Here, when the accuracy of the calculated response is less than or equal to the threshold level for the activation rate, the response may be derived through the natural language processing engine set as the second processing engine.

For another example, in the case of a natural language processing model composed of a simultaneous linkage processing model, the chatbot may derive a response by simultaneously using a similar sentence-based natural language processing engine and an intention-based natural language processing engine. When the simultaneous linkage processing model is selected, a response is derived through each of the natural language processing engine set as the first processing engine and the natural language processing engine set as the second processing engine, and the reliability of each derived response is calculated. Of each response derived based on the reliability, the final response can be determined. Here, the response derivation unit 260 may assign a predetermined weight to the reliability of each derived response, and determine a final response among the derived responses based on the reliability of each response to which the weight is assigned.

When one of a plurality of services registered in a messenger app providing an instant messaging service is selected by the user of the user terminal 130, the communication unit 270 transmits service account information corresponding to the selected service to the messenger server. It can be received from 120.

For example, when a service provided by a specific official account among a plurality of services is selected by the user of the user terminal 130, the communication unit 270 sends the official account ID or chatbot ID corresponding to the selected service to the messenger server 120 And the communication unit 270 may receive an official account ID or chatbot ID from the messenger server 120.

For another example, when a service provided by a specific chatbot among a plurality of services is selected by the user of the user terminal 130, the communication unit 270 transmits the chatbot ID corresponding to the selected service to the messenger server 120, The communication unit 270 may receive the chatbot ID from the messenger server 120. Here, the service provided by the chatbot may include, for example, a weather service, a store service, and a shopping mall service. For example, when the user of the user terminal 130 selects a weather service provided by a chatbot from among a plurality of services registered in the messenger app, the communication unit 270 sends a chatbot ID corresponding to the weather service from the messenger server 120. Can receive.

In addition, the communication unit 270 may further receive the user ID of the user terminal 130 from the messenger server 120.

The communication unit 270 may receive a query message input in relation to a service selected by a user from the messenger server 120. Here, the query message may be composed of a short sentence or a long sentence, and the length of the sentence may not be limited. For example, when the user selects the weather service, a query message related to weather is transmitted from the user terminal 130 to the message server 120, and the communication unit 270 receives the query message from the messenger server 120. I can.

The communication unit 270 may transmit a response message to the query message of the user terminal 130 derived from the chatbot to the messenger server 120. In this case, the response message may be transmitted from the messenger server 120 to the user terminal 130.

5A to 5C are exemplary diagrams for explaining a process of setting a threshold level for an activation rate according to an embodiment of the present invention.

5A is an exemplary diagram illustrating a process of setting an activation rate through a user interface for a bot builder according to an embodiment of the present invention. Referring to FIG. 5A, the management server 110 may receive a setting button 502 on the knowledge extension page 501 of the'knowledge +'menu 500 from the administrator through the user interface for the bot builder.

The management server 110 may receive at least one of a plurality of natural language processing models built on the basis of at least one of an intention-based natural language processing engine and a similar sentence natural language processing engine from an administrator. In this case, the management server 110 may receive a threshold for an activation rate that determines a processing rate between at least two natural language processing engines from an administrator. Here, the threshold for the activation rate may play a role of determining the processing order of each natural language processing engine. Hereinafter, an embodiment in which a threshold for an activation rate is set based on a sequential linkage processing model and a simultaneous linkage processing model in which a similar sentence-based natural language processing engine and an intention-based natural language processing engine are used in combination will be described.

5B is an exemplary diagram for explaining a process of adjusting a bar displayed on an activation gauge according to an embodiment of the present invention. Referring to FIG. 5B, when the setting button 502 is input on the knowledge expansion page 501, the management server 110 may display a pop-up window 510 for adjusting the activation gauge.

The administrator can set the threshold level for the activation rate by dragging and adjusting the bars 514 displayed on the activation gauge 511 through the pop-up window 510 for adjusting the activation gauge. Here, in the case of a natural language processing model in which an intention-based natural language processing engine and a similar sentence-based natural language processing engine are combined, the threshold level for determining the activation rate for each natural language processing engine is 0.0 (lowest point) and 1.0 (highest point) in units of 0.1. Can be adjusted between.

For example, an activation rate of 0% (512) means an off setting of an operation for a similar sentence-based natural language processing engine, and this means that the level for the activation rate becomes 0.0, so that only the intention-based natural language processing engine is used. The user's intention can be extracted from the query message, and a response corresponding to the extracted user's intention can be provided. In contrast, the activation rate of 100% (513) means an off setting of the operation for the intention-based natural language processing engine, and this means that the level for the activation rate is 1.0 (the function of the similar sentence-based natural language processing engine is strongly operated). Accordingly, it is possible to provide a response to a query message using only a similar sentence-based natural language processing engine based on upload data composed of query data and response data corresponding to the query data.

That is, when the threshold level for the activation rate is set for a natural language processing model (for example, an intention-based natural language processing engine -> a sequential linkage processing model in which the similar sentence natural language processing engine operates in sequence), the intention-based natural language processing engine A response may be derived through an intention-based natural language processing engine or a similar sentence-based natural language processing engine according to whether the derived response is below the threshold level.

For example, if the accuracy of the response derived to the query message through the intention-based natural language processing engine is greater than the threshold (response accuracy> threshold), the similar sentence-based natural language processing engine is bypassed, and intention-based natural language processing You can make sure that the response derived by the engine is provided.

As another example, when the accuracy of the response derived to the query message through the intent-based natural language processing engine is equal to or less than the threshold (response accuracy ≤ threshold), the intention-based natural language processing engine is bypassed, and similar sentences The response derived by the underlying natural language processing engine can be provided.

At this time, when the administrator adjusts the activation rate lever upward, the probability that a response is provided by the intention-based natural language processing engine increases as the threshold value increases closer to 1.0, and if the activation rate lever is adjusted downward, the threshold value increases. As it goes down closer to 0.0, the probability of providing a response by a similar sentence-based natural language processing engine may increase.

In addition, when machine learning is applied to an intention-based natural language processing engine, a result value of machine learning for a user's speech may be filtered while adjusting a threshold level for an activation rate between 0 and 1.

6A to 6E are exemplary diagrams for explaining a process of setting a derived response to be output in an immediate response type according to an embodiment of the present invention.

Referring to FIG. 6A, the management server 110 may receive an output setting 600 of a response to be output from the user terminal 130 from an administrator. For example, the management server 110 may receive an immediate immediate response button 601 from an administrator through the output setting 600. In the case of an immediate immediate response type, for example, when the management server 110 receives a query message "Tell me how to return" 602 from the user terminal 130, the management server 110 says "Yes, the return procedure It is possible to output a speech bubble containing a single response corresponding to the query message, such as "".

Referring to FIG. 6B, the management server 110 may receive a guide speech balloon 611 set to be displayed as text 612 from an administrator through an immediate response type setting 610. For example, before the response is output from the user terminal 130, the management server 110 may be set to display a text guide phrase such as "An answer related to your inquiry is as follows!" In addition, the management server 110 may receive a setting for on/off of the feedback button 614, such as'like' or'bad', from an administrator.

Referring to FIG. 6C, the management server 110 may receive a guide speech balloon 611 set to be displayed as an image 613 from an administrator through an immediate response type setting 610. For example, the management server 110 may be set to display a company logo in an image format or a chatbot character before a response is output from the user terminal 130.

7A to 7C are exemplary diagrams for explaining a process of setting a derived response to be output in an answer recommendation type according to an embodiment of the present invention.

Referring to FIG. 7A, the management server 110 may receive an output setting 700 to be output from the user terminal 130 from an administrator. For example, the management server 110 may receive an answer recommendation type button 701 from an administrator through the output setting 700.

In the case of the answer recommendation type, for example, when the management server 110 receives a query message saying "Tell me how to return" 702 from the user terminal 130, the management server 110 asks, If you choose, I will respond to you in the future.", along with the guide phrase (703), "Q1) Please tell me the details of the return procedure. Q2) Please tell me the return item. Q3) Tell me the items excluded from return." The sentence 704 and a response corresponding to each query sentence may be provided to the user terminal 130 through the messenger server 120. At this time, in the answer recommendation type, when any one of the predetermined similar query sentences 704 is selected from the user terminal 130, a response corresponding to the selected similar query sentence among the plurality of responses 705 is the user terminal 130 Can be provided as

Referring to FIG. 7B, the management server 110 may receive a guide speech balloon 711 set to be displayed as text 712 from an administrator through the answer recommendation type setting 710. For example, before the response is output from the user terminal 130, the management server 110 may be set to display a text guide phrase such as "An answer related to your inquiry is as follows!"

Referring to FIG. 7C, the management server 110 registers the upload data in which the image URL information 722 is inserted along with the question data 720 and the response data 721 corresponding to the question data 720. , The user terminal 130 may display, for example, a similar query sentence, a response corresponding to the similar query sentence, and a guide speech bubble 730 including an image automatically converted from a corresponding URL. In this case, when image URL information does not exist, an image uploaded by default (eg, a company logo, a common guide image, etc.) may be displayed.

Referring to FIG. 7D, the management server 110 may allow the guide speech bubble 711 to be displayed as an image 740 through the answer recommendation type setting 710 from the administrator. In this case, the user terminal 130 may display, for example, an image converted from a corresponding URL, a similar query message, and a guide speech bubble including a response message to the similar query message.

Referring to FIG. 7E, when an administrator registers upload data into which web landing URL information is inserted along with question data and response data corresponding to the question data, the management server 110 is the user terminal 130 ), when the response is output, the link can be automatically inserted into the message as a separate action inducing button. In this case, the user terminal 130 may display, for example, a similar query sentence, a response corresponding to the similar query sentence, and a speech balloon including “go to link”.

8A and 8B are flowcharts of a method for deriving a response from a management server according to an embodiment of the present invention.

8A is a flowchart of a method for deriving a response using a natural language processing engine in a management server according to an embodiment of the present invention.

Referring to FIG. 8A, in step S810, the management server 110 may register upload data consisting of question data and response data corresponding to the question data.

In step S820, the management server 110 may train a similar sentence-based natural language processing engine based on the registered upload data.

In step S830, the management server 110 may generate a chatbot using the generated similar sentence-based natural language processing engine.

In step S840, the management server 110 may derive a response to the query message using a similar sentence-based natural language processing engine through the generated chatbot.

In this way, the chatbot can derive a response to a query message using a single natural language processing engine.

According to an embodiment, the chatbot may derive a response to a query message based on a knowledge search using a natural language processing engine based on a similar sentence. For example, the chatbot may search for a plurality of similar query sentences from the query message. In this case, the chatbot may extract a predetermined similar query sentence having high similarity among a plurality of similar query sentences, and derive the extracted predetermined similar query sentence and a response corresponding thereto.

According to another embodiment, the chatbot may derive a response to a query message based on machine learning using an intention-based natural language processing engine. For example, the chatbot may extract a user's intention based on an entity included in a query message and derive a response corresponding to the extracted user's intention.

8B is a flowchart of a method for deriving a response using a natural language processing model in a management server according to an embodiment of the present invention.

Referring to FIG. 8B, in step S850, the management server 110 may build a plurality of natural language processing models based on at least one of a similar sentence-based natural language processing engine and an intention-based natural language processing engine.

In step S860, the management server 110 may select at least one of a plurality of constructed natural language processing models.

In step S860, the management server 110 may generate a chatbot using the selected natural language processing model.

In step S870, the management server 110 may derive a response to the query message using the natural language processing model selected through the generated chatbot.

In the above description, steps S810 to S870 may be further divided into additional steps or combined into fewer steps, according to an embodiment of the present invention. In addition, some steps may be omitted as necessary, and the order between steps may be switched.

9A and 9B are exemplary diagrams for explaining a process of deriving a response to a query message by using a natural language processing model in a management server according to an embodiment of the present invention. Here, the natural language processing model may be composed of a combination of an intention-based natural language processing engine and a similar sentence-based natural language processing engine, and by receiving a threshold level for the activation rate through an activation gauge by an administrator, each natural language processing engine The activation rate (degree of interference) can be adjusted.

A chatbot using such a natural language processing model can operate smoothly in an environment of a natural language processing engine composed of dual, triple, and multi.

9A is an exemplary diagram for explaining a process of deriving a response through a chatbot using a sequential linkage processing model in a management server according to an embodiment of the present invention. Referring to FIG. 9A, the management server 110 may derive a response to a query message through a chatbot using a sequential linking processing model so that the first processing engine and the second processing engine are sequentially performed in a preset order. .

Hereinafter, it is assumed that the first processing engine is an intention-based natural language processing engine, and the second processing engine is a similar sentence-based natural language processing engine. Of course, the first processing engine may be set as the similar sentence-based natural language processing engine, and the second processing engine may be set as the intention-based natural language processing engine by different processing orders between natural language processing engines according to the administrator's setting. At this time, the threshold level for the activation rate that causes the first processing engine to be performed may be adjusted by the administrator between 0 and 1, and flow into the second processing engine through filtering of the result value derived through the first processing engine. The range can be increased or decreased through adjustment of the threshold level. The threshold for the activation rate between the first processing engine 910 and the second processing engine 920 may be set by an administrator as described through FIGS. 5A to 5C.

When receiving the query message 900 input from the user terminal 130 from the messenger server 120, the management server 110 extracts the user intention for the query message 900 through the first processing engine 910 , A first response may be derived (911) based on the extracted user intention. At this time, when the similarity of the first response is less than the threshold 912 level for the preset activation rate, the management server 110 fails to extract the user intention through the first processing engine 910 It is determined 913 to search for a plurality of similar sentences for the query message 900 through the second processing engine 920, and a second response may be derived 921 based on the searched plurality of similar sentences. . In this case, if no response is retrieved (922), the management server 110 fallback (Fallback, 930) is connected to the platform API 932 No of responses (931) or third parties (3 ^rd -party) through You can do it.

9B is an exemplary diagram for explaining a process of deriving a response through a chatbot using a simultaneous linkage processing model in a management server according to an embodiment of the present invention. Referring to FIG. 9B, the management server 110 may derive a response to a query message through a chatbot using a simultaneous link processing model in which the first processing engine and the second processing engine are simultaneously performed.

Hereinafter, it is assumed that the first processing engine is an intention-based natural language processing engine, and the second processing engine is a similar sentence-based natural language processing engine.

When the management server 110 receives the query message 900 input from the user terminal 130 from the messenger server 120, the first processing engine 910 and the second processing engine 920 The reliability of each derived response can be calculated. For example, the management server 110 calculates a first reliability score 940 for a response derived through the first processing engine 910, and provides a first reliability score 940 for the response derived through the second processing engine 920. 2 A reliability score 950 may be calculated.

The management server 110 may correct (960) the reliability of each response by assigning a predetermined weight level to the calculated reliability of each response. In addition, the management server 110 may control the performance of a specific natural language processing engine to stand out by adjusting the weight level to be given between the reliability of each response.

The management server 110 may compare the corrected reliability score of each response to derive a response with a high score (970), and determine the derived response as a final response (980).

The present invention proposes that the threshold for the activation rate can be customized by the administrator, so that when the natural language processing model is composed of the first natural language processing engine and the second natural language processing engine as a sequentially linked model, each natural language processing engine operates. You can have bullets created. In addition, the present invention proposes that the weight level to be assigned between the reliability of responses derived by each natural language processing engine can be customized by the administrator, so that the natural language processing model is simultaneously linked with the first natural language processing engine and the second natural language processing engine. When configured as a model, a first natural language processing engine and a second natural language processing engine may be simultaneously operated, and a reference point for comparison and determination may be generated by assigning a weight to the accuracy calculated by each natural language processing engine.

An embodiment of the present invention may also be implemented in the form of a recording medium including instructions executable by a computer, such as a program module executed by a computer. Computer-readable media can be any available media that can be accessed by a computer, and includes both volatile and nonvolatile media, removable and non-removable media. Further, the computer-readable medium may include all computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.

Although the methods and systems of the present invention have been described in connection with specific embodiments, some or all of their components or operations may be implemented using a computer system having a general-purpose hardware architecture.

The above description of the present invention is for illustrative purposes only, and those of ordinary skill in the art to which the present invention pertains will be able to understand that other specific forms can be easily modified without changing the technical spirit or essential features of the present invention will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not limiting. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as being distributed may also be implemented in a combined form.

The scope of the present invention is indicated by the claims to be described later rather than the detailed description, and all changes or modified forms derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention. do.

110: management server
120: messenger server
130: user terminal
210: Upload data registration
220: Similar sentence-based natural language processing engine learning unit
230: Natural language processing model management unit
240: setting unit
250: chatbot generation unit
260: response derivation unit
270: communication department

Claims (18)

In the method of managing the natural language processing engine,
Registering upload data consisting of question data and response data corresponding to the question data;
Performing an indexing process on the registered uploaded data;
Learning a similar sentence-based natural language processing engine based on the uploaded data on which the indexing process has been performed;
Generating a chatbot using the generated similar sentence-based natural language processing engine; And
Deriving a response to the query message from the uploaded data on which the indexing process has been performed using the similar sentence-based natural language processing engine through the generated chatbot
That includes, a natural language processing engine management method.
The method of claim 1,
The step of receiving the upload data is registered,
The method of managing a natural language processing engine to receive the uploaded data through a file upload method or an API connection method with a management server.
The method of claim 1,
Receiving at least one selected from among a plurality of previously registered upload data; And
Learning a similar sentence-based natural language processing engine based on the selected upload data
That further comprises a natural language processing engine management method.
The method of claim 1,
Constructing a plurality of natural language processing models based on at least one of the similar sentence-based natural language processing engine and the intention-based natural language processing engine;
Receiving at least one of the constructed natural language processing models;
Generating a chatbot using the selected natural language processing model; And
Deriving a response to the query message using the selected natural language processing model through the generated chatbot
That further comprises a natural language processing engine management method.
The method of claim 4,
The natural language processing model includes a sequential linkage processing model,
The sequential linkage processing model is to derive the response by sequentially using the similar sentence-based natural language processing engine and the intention-based natural language processing engine in a preset order.
The method of claim 5,
If the sequentially linked processing model is selected, deriving the response through a natural language processing engine set as a first processing engine; And
When the response is not derived through the first processing engine, deriving the response through a natural language processing engine set as a second processing engine
That further comprises a natural language processing engine management method.
The method of claim 6,
Calculating an accuracy of a response derived through a natural language processing engine set as the first processing engine; And
When the accuracy of the calculated response exceeds a threshold level for an activation rate preset between the first processing engine and the second processing engine, determining the derived response as a final response
That further comprises a natural language processing engine management method.
The method of claim 7,
When the accuracy of the calculated response is less than or equal to the threshold level for the activation rate, deriving the response through a natural language processing engine set as the second processing engine
That further comprises a natural language processing engine management method.
The method of claim 7,
Step of receiving the threshold level set by adjusting a bar displayed on an activation gauge
That further comprises a natural language processing engine management method.
The method of claim 4,
The natural language processing model includes a simultaneous linkage processing model,
The simultaneous linkage processing model is to derive the response by simultaneously using the similar sentence-based natural language processing engine and the intention-based natural language processing engine.
The method of claim 10,
When the simultaneous link processing model is selected, deriving the response through each of a natural language processing engine set as a first processing engine and a natural language processing engine set as a second processing engine;
Calculating the reliability of each of the derived responses; And
Determining a final response among the derived responses based on the calculated reliability of each response
That further comprises a natural language processing engine management method.
The method of claim 11,
Assigning a predetermined weight to the reliability of each of the derived responses; And
Determining a final response among the derived responses based on the reliability of each response to which the weight is added
That further comprises a natural language processing engine management method.
The method of claim 1,
Receiving an output type of the response set as an answer recommendation type;
Deriving a plurality of responses based on the similarity calculated using the similar sentence-based natural language processing engine;
Providing the derived plurality of responses to the user terminal that generated the query message
That further comprises a natural language processing engine management method.
In the management server for managing the natural language processing engine,
An upload data registration unit for registering upload data consisting of question data and response data corresponding to the question data;
An indexing unit for performing an indexing process on the registered uploaded data;
A similar sentence-based natural language processing engine learning unit for learning a similar sentence-based natural language processing engine based on the uploaded data on which the indexing process has been performed;
A chatbot generator that generates a chatbot using the generated similar sentence-based natural language processing engine; And
Response derivation unit for deriving a response to the query message from the uploaded data on which the indexing process has been performed using the similar sentence-based natural language processing engine through the generated chatbot
That containing, the management server.
The method of claim 14,
A natural language processing model management unit configured to construct a plurality of natural language processing models based on at least one or more of the similar sentence-based natural language processing engine and the intention-based natural language processing engine, and to select at least one of the built-up natural language processing models. and,
The chatbot generation unit generates a chatbot using the selected natural language processing model,
The response derivation unit is to derive a response to the query message by using the selected natural language processing model through the generated chatbot.
The method of claim 15,
The natural language processing model includes a sequential linkage processing model,
The sequential linkage processing model is to derive the response by sequentially using the similar sentence-based natural language processing engine and the intention-based natural language processing engine in a preset order.
The method of claim 15,
The natural language processing model includes a simultaneous linkage processing model,
The simultaneous linkage processing model is to derive the response by simultaneously using the similar sentence-based natural language processing engine and the intention-based natural language processing engine.
A computer program stored in a computer readable medium comprising a sequence of instructions for managing a natural language processing engine,
When the computer program is executed by a computing device,
Receive upload data consisting of question data and response data corresponding to the question data,
Performing an indexing process on the registered upload data,
Learning a natural language processing engine based on similar sentences based on the uploaded data on which the indexing process has been performed,
Create a chatbot using the generated similar sentence-based natural language processing engine,
A computer program stored in a medium comprising a sequence of instructions for deriving a response to a query message from uploaded data on which the indexing process has been performed using the similar sentence-based natural language processing engine through the generated chatbot.

Images